← Clinical Trials
Head & Neck Cancer Phase 3 Overall Survival NCT02252042

KEYNOTE-040: Complete Statistical Analysis of Pembrolizumab in Recurrent or Metastatic Head and Neck Cancer

An independent statistical analysis of the randomized phase 3 KEYNOTE-040 trial comparing pembrolizumab with standard treatment in participants with recurrent or metastatic head and neck squamous cell cancer.

Trial status: Completed  ·  Enrollment: 495  ·  Primary completion: 15-May-2017
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

KEYNOTE-040 was a randomized, parallel-group, open-label phase 3 trial evaluating pembrolizumab versus standard treatment in participants with recurrent or metastatic head and neck squamous cell cancer. The registry reports 495 enrolled participants, two arms, two registered primary endpoints, and formal statistical analyses for both primary endpoints.

495
Enrolled
2 treatment arms
2
Primary endpoints
Both time-to-event
0.80
Updated OS HR
95% CI 0.65–0.98
0.01605
Updated OS P-value
Two-sided analysis
FeatureKEYNOTE-040
PhasePhase 3
ConditionHead and Neck Squamous Cell Cancer
DesignRandomized, parallel
MaskingNone
AllocationRandomized
Primary purposeTreatment
Enrollment495
Primary endpointsInitial Overall Survival (OS) for All Participants; Updated Final OS for All Participants
Results postedYes
Outcome measures posted17
Statistical analyses posted11
Lead sponsorMerck Sharp & Dohme LLC
Sponsor typeIndustry
ClinicalTrials.govNCT02252042

2. Clinical Question

The central statistical question was whether pembrolizumab produced a different time-to-event outcome from standard treatment in participants with recurrent or metastatic head and neck squamous cell cancer. The registry classifies the primary hypothesis type as superiority.

Population

Participants with recurrent or metastatic head and neck squamous cell cancer enrolled in the randomized phase 3 study.

Intervention

Pembrolizumab, identified in the registry as a biological intervention.

Comparator

Standard treatment. The registry identifies methotrexate, docetaxel, and cetuximab as the drug and biological interventions associated with the standard-treatment arm.

Primary question

Does pembrolizumab improve overall survival relative to standard treatment under the prespecified superiority framework?

3. Trial Design

01
Enroll495 participants
02
Randomize2 treatment arms
03
TreatPembrolizumab or standard treatment
04
FollowTime-to-event outcomes
05
AnalyzeLog-rank / Cox methods
ARM 1

Pembrolizumab

  • Biological intervention
  • Compared with standard treatment
  • Included in the randomized efficacy population according to assigned treatment
ARM 2

Standard Treatment

  • Comparator treatment
  • Registry-listed interventions include methotrexate, docetaxel, and cetuximab
  • Included in the randomized efficacy population according to assigned treatment

The study was open-label rather than masked. Its allocation was randomized and its design model was parallel. The efficacy analyses reported in the registry use all randomized participants and retain participants in the treatment group to which they were randomized.

4. Trial Timeline

17-Nov-2014

Study start

The registry lists 17-Nov-2014 as the study start date.

15-May-2017

Primary completion

The registry lists 15-May-2017 as the primary completion date. The initial OS analysis uses a data cutoff of 15-May-2017.

04-Jun-2017

Database lock for initial OS results

The registered definition for the initial OS analysis identifies 04-Jun-2017 as the database lock date.

Completed

Registry status

ClinicalTrials.gov lists the study status as completed.

5. Endpoints

The registry identifies two primary endpoints, both classified as time-to-event outcomes. Both are overall-survival analyses for all participants, with an initial analysis and an updated final analysis.

EndpointTime frameDefinition / statistical role
Initial Overall Survival (OS) for All Participants Up to approximately 2 years OS was defined as the time from randomization to death due to any cause. Participants without documented death at the time of the final analysis were to be censored at the date of the last follow-up. The initial OS results use a data cutoff date of 15-May-2017 and database lock date of 04-Jun-2017.
Updated Final OS for All Participants Up to approximately 2 years OS was defined as the time from randomization to death due to any cause. Participants without documented death at the time of the final analysis were to be censored at the date of the last follow-up. The updated OS analysis used complete acquisition of outstanding survival data with a 15-May-2017 data cutoff.

Secondary endpoints with posted analyses

The registry also reports formal analyses for OS in participants with PD-L1 ≥1% CPS, several progression-related endpoints, and objective response rate. The posted analyses use hazard ratios for time-to-event outcomes and differences in percentages for response outcomes.

Secondary endpointTime frameEffect measure
OS in participants with PD-L1 ≥1% CPSUp to approximately 2 yearsHazard ratio
PFS per RECIST 1.1 for all participantsUp to approximately 2 yearsHazard ratio
PFS per RECIST 1.1 in participants with PD-L1 ≥1% CPSUp to approximately 2 yearsHazard ratio
ORR per RECIST 1.1 for all participantsUp to approximately 2 yearsDifference in percentages
ORR per RECIST 1.1 in participants with PD-L1 ≥1% CPSUp to approximately 2 yearsDifference in percentages
TTP per RECIST 1.1 for all participantsUp to approximately 2 yearsHazard ratio
TTP per RECIST 1.1 in participants with PD-L1 ≥1% CPSUp to approximately 2 yearsHazard ratio
PFS per modified RECIST for all participantsUp to approximately 2 yearsHazard ratio
PFS per modified RECIST 1.1 in participants with PD-L1 ≥1% CPSUp to approximately 2 yearsHazard ratio

6. Statistical Methodology

Log-rank testing

The registry identifies the log-rank test as the reported method for the posted analyses. The log-rank test is designed to compare survival distributions between treatment groups while accounting for the timing of events and right-censored observations.

Time-to-event comparison
H0: survival distributions are equal

For a superiority analysis, the treatment comparison asks whether the observed event-time distributions differ between randomized groups. The log-rank test uses the ordering and timing of events rather than reducing follow-up to a single binary outcome.

Cox regression

The registry analysis notes specify Cox regression models for the hazard-ratio analyses. For the initial OS analysis, treatment was included as a covariate and the model was stratified by ECOG performance status, HPV status, and PD-L1 status. The same stratification factors are specified for the updated final OS analysis, which the registry labels a nominal HR analysis.

Hazard ratio
HR = estimated hazard in pembrolizumab group ÷ estimated hazard in standard-treatment group

An HR below 1 indicates a lower estimated instantaneous event rate in the pembrolizumab group under the fitted model. It is a relative time-to-event measure, not an absolute probability and not the percentage of participants who benefit.

Stratified analysis

For the primary OS analyses, the Cox regression model was stratified by ECOG PS (0 vs. 1), HPV status (Positive vs. Negative), and PD-L1 status (Strongly Positive, Not Strongly Positive). Stratification allows the baseline hazard to differ across these strata while estimating the treatment effect within the specified model framework.

Covariate adjustment

The registry explicitly identifies covariate adjustment and stratified analysis as concepts in the primary analyses. The PD-L1-positive PFS analysis is an example where the analysis notes specify a Cox regression model with treatment as a single covariate rather than the three-factor stratification used for the primary OS analyses.

Analysis population

For every posted efficacy analysis reported in the ClinicalTrials.gov record, the efficacy population consists of all randomized participants for the all-participant analyses. Participants are included in the treatment group to which they were randomized. The PD-L1-positive analyses restrict that randomized efficacy population to participants with PD-L1 ≥1% CPS.

7. Primary Results: Initial Overall Survival

The first registered primary endpoint is Initial Overall Survival (OS) for All Participants. The endpoint was analyzed through approximately 2 years, with OS defined from randomization to death from any cause and censoring at last follow-up for participants without documented death at the time of final analysis.

Hazard ratio for overall survival

0.82

95% CI: 0.67–1.01   ·   P = 0.03160

Two-sided confidence interval · Superiority hypothesis

FeatureInitial OS analysis
PopulationAll randomized participants
ComparisonPembrolizumab vs Standard Treatment
MethodLog-rank test; Cox regression for HR
Effect measureHazard ratio
Estimate0.82
95% CI0.67–1.01
P-value0.03160
ModelCox regression with treatment as a covariate, stratified by ECOG PS, HPV status, and PD-L1 status
Clinical Biostats interpretation

The estimated hazard ratio of 0.82 means that the fitted analysis estimated the instantaneous rate of death in the pembrolizumab group at approximately 82% of the rate in the standard-treatment group, corresponding to a 18% lower estimated hazard under that model.

The HR does not mean that 18% of participants benefited, that 18% fewer participants died, or that an individual participant's probability of death was reduced by exactly 18%. Hazard ratios summarize relative event rates over time and depend on the underlying time-to-event model.

The 95% CI of 0.67–1.01 describes statistical uncertainty around the estimated hazard ratio. Its upper endpoint is close to 1 and slightly above it, so the interval indicates that the estimate is not highly precise. A confidence interval is not a range containing the true effect with 95% probability; it is an interval produced by the specified statistical procedure under its repeated-sampling interpretation.

The p-value of 0.03160 addresses the compatibility of the observed data with the specified null hypothesis under the statistical test. It does not measure the size or clinical importance of the treatment effect. The effect size is communicated by the HR, while its uncertainty is communicated by the confidence interval.

Because this is a time-to-event analysis, censoring and the assumptions underlying the Cox model matter. The registry reports a stratified Cox model but does not provide, in the ClinicalTrials.gov record, a formal assessment of the proportional-hazards assumption.

8. Primary Results: Updated Final Overall Survival

The second registered primary endpoint is the Updated Final OS for All Participants. The registry describes this as an updated analysis after complete acquisition of outstanding survival data using a 15-May-2017 data cutoff.

Updated final hazard ratio for overall survival

0.80

95% CI: 0.65–0.98   ·   P = 0.01605

Two-sided confidence interval · Superiority hypothesis · Nominal HR

FeatureUpdated final OS analysis
PopulationAll randomized participants
ComparisonPembrolizumab vs Standard Treatment
MethodLog-rank test; Cox regression for HR
Effect measureHazard ratio
Estimate0.80
95% CI0.65–0.98
P-value0.01605
ModelCox regression with treatment as a covariate, stratified by ECOG PS, HPV status, and PD-L1 status
Clinical Biostats interpretation

The updated HR of 0.80 corresponds to a 20% lower estimated hazard of death for pembrolizumab relative to standard treatment under the fitted Cox model. This is a relative instantaneous event-rate interpretation, not a statement that 20% of participants avoided death or that each participant experienced the same proportional reduction.

The 95% CI of 0.65–0.98 gives the statistical uncertainty around the estimated HR. Unlike the initial analysis, the registry-reported updated interval lies below 1.00 at its upper boundary, although the interval still permits a range of treatment-effect magnitudes.

The p-value of 0.01605 describes the evidence against the specified null hypothesis under the reported analysis. It is not a measure of effect size. The HR communicates the estimated relative effect, and the CI communicates its precision.

The registry describes this as a nominal HR. That wording is important: the numerical estimate should be interpreted in the context of the analysis framework and any multiplicity considerations rather than treating the p-value as a standalone measure of clinical importance.

The primary analysis uses all randomized participants according to assigned treatment. That preserves the randomized comparison for efficacy, while censoring and the proportional-hazards framework remain important considerations in interpreting a single HR.

9. Secondary Results: Overall Survival in Participants With PD-L1 ≥1% CPS

The registry reports a secondary OS analysis restricted to participants with PD-L1 ≥1% CPS. Participants remained classified according to their randomized treatment group.

OS hazard ratio in PD-L1 ≥1% CPS

0.74

95% CI: 0.58–0.93   ·   P = 0.00493

The Cox model included treatment as a covariate and was stratified by ECOG PS, HPV status, and PD-L1 status. The estimated HR of 0.74 corresponds to a 26% lower estimated instantaneous hazard of death for pembrolizumab under the fitted model. The confidence interval provides the uncertainty around that estimate; the p-value addresses the hypothesis test and does not quantify effect magnitude.

10. Secondary Results: Progression-Free Survival

EndpointHR95% CIP-value
PFS per RECIST 1.1, all participants 0.96 0.79–1.16 0.32504
PFS per RECIST 1.1, PD-L1 ≥1% CPS 0.86 0.69–1.06 0.07736
PFS per modified RECIST, all participants 1.04 0.86–1.27 0.65759
PFS per modified RECIST 1.1, PD-L1 ≥1% CPS 1.01 0.81–1.26 0.51982

All four analyses were reported as time-to-event analyses using log-rank testing and Cox regression. The all-participant RECIST 1.1 analysis produced an HR of 0.96, meaning the estimated hazard was close to the comparator group's hazard under the fitted model. Its 95% CI of 0.79–1.16 spans 1, and the reported p-value was 0.32504.

In the PD-L1 ≥1% CPS population, the RECIST 1.1 PFS HR was 0.86 with a 95% CI of 0.69–1.06 and P = 0.07736. The modified RECIST analyses produced HRs of 1.04 and 1.01 for the all-participant and PD-L1 ≥1% CPS populations, respectively.

How to read these PFS results

The estimates should be read together with their confidence intervals rather than by p-value alone. An HR near 1 indicates little estimated relative difference in instantaneous progression-or-death rates under the fitted model. An interval crossing 1 indicates that the data are compatible with treatment effects on both sides of the null value at the stated confidence level.

The fact that several PFS analyses use different response definitions or analysis populations also matters. They should not be collapsed into one numerical conclusion, because the endpoint definitions and populations are not identical.

11. Secondary Results: Objective Response Rate

EndpointRisk difference95% CIP-value
ORR per RECIST 1.1, all participants 4.6 percentage points -1.2 to 10.6 0.0610
ORR per RECIST 1.1, PD-L1 ≥1% CPS 7.5 percentage points 0.6 to 14.6 0.0171

The registry reports the effect measure as a difference in percentages, normalized as a risk difference. For the all-participant analysis, the estimate was 4.6 percentage points with a 95% CI of -1.2 to 10.6 and P = 0.0610. For participants with PD-L1 ≥1% CPS, the estimate was 7.5 percentage points with a 95% CI of 0.6 to 14.6 and P = 0.0171.

Risk difference interpretation

A risk difference expresses an absolute difference in the percentage experiencing the response outcome between groups. An estimate of 7.5 means a 7.5-percentage-point difference under the registry's analysis definition; it is not a relative percentage increase.

The confidence interval is particularly useful here because it expresses the uncertainty in absolute terms. For the all-participant analysis, the interval extends from -1.2 to 10.6 percentage points, so the estimated effect is not highly precise. For the PD-L1 ≥1% CPS analysis, the interval extends from 0.6 to 14.6 percentage points.

The p-value tests the stated null hypothesis of a zero percentage difference. It does not tell the reader whether a 7.5-percentage-point difference is clinically important; that is a separate question requiring clinical context.

12. Secondary Results: Time to Progression

EndpointHR95% CIP-value
TTP per RECIST 1.1, all participants 0.89 0.70–1.12 0.14545
TTP per RECIST 1.1, PD-L1 ≥1% CPS 0.81 0.62–1.06 0.05851

The all-participant TTP analysis produced an HR of 0.89 with a 95% CI of 0.70–1.12 and P = 0.14545. In the PD-L1 ≥1% CPS population, the HR was 0.81 with a 95% CI of 0.62–1.06 and P = 0.05851.

These are time-to-event analyses, so the HR summarizes the relative instantaneous rate of the defined progression event. As with OS and PFS, the confidence interval should be considered alongside the point estimate, and a p-value should not be interpreted as a measure of the magnitude of the treatment effect.

13. Statistical Methods Explained

Why was a log-rank test used?

The registry identifies the log-rank test as the reported method for the time-to-event analyses. The test compares survival distributions while using the timing of events and accommodating right-censored observations. This makes it appropriate for endpoints such as OS, PFS, and TTP rather than treating follow-up as a simple yes/no outcome at a fixed date.

What does an OS hazard ratio of 0.80 mean?

An HR of 0.80 means that the fitted Cox model estimated the instantaneous death rate in the pembrolizumab group at 80% of that in the standard-treatment group over the analyzed follow-up. It can be described as a 20% lower estimated hazard, but it should not be translated directly into a 20% lower probability of death for every participant.

Why was stratification used in the primary OS model?

The registry specifies stratification by ECOG PS, HPV status, and PD-L1 status. Stratification allows the baseline event process to differ across these predefined groups while the treatment effect is estimated within the Cox framework. It therefore accounts for these factors without requiring a single common baseline hazard across all strata.

Why does the confidence interval matter as much as the HR?

The HR is a point estimate, while the confidence interval shows the uncertainty associated with that estimate under the statistical model. For the updated OS analysis, the HR was 0.80 and the 95% CI was 0.65–0.98. The interval communicates considerably more information about precision than the point estimate alone.

Why does the p-value not measure treatment effect size?

A p-value quantifies the compatibility of the observed data with a specified null hypothesis under the test procedure. It depends on both the observed effect and the amount of information in the data. Two studies can have the same effect estimate but different p-values if their precision differs. Effect magnitude should therefore be described using measures such as the HR or risk difference, alongside the confidence interval.

Why are the all-participant and PD-L1 ≥1% CPS analyses different populations?

The all-participant analyses include all randomized participants, whereas the PD-L1 ≥1% CPS analyses restrict the efficacy population to randomized participants meeting that biomarker definition. Consequently, the estimates answer different statistical questions and should not be treated as interchangeable.

14. Randomization and Analysis Population

Randomization is the foundation of the treatment comparison. By assigning participants to treatment groups before outcome assessment, randomized allocation provides the framework for comparing outcomes between groups while reducing systematic differences attributable to treatment assignment.

Allocation
Randomized
Design model
Parallel
Masking
None
Primary purpose
Treatment

The primary efficacy analyses retain participants in the treatment group to which they were randomized. This is an important feature of the analysis because switching participants between treatment groups based on treatment received can undermine the original randomized comparison.

15. Stratified Analysis and Covariate Adjustment

The primary OS analyses provide a useful example of the difference between an unadjusted treatment comparison and a model incorporating prespecified structure. The registry reports Cox regression with treatment as a covariate and stratification by three baseline factors.

Stratification factorReported categories
ECOG performance status0 vs. 1
HPV statusPositive vs. Negative
PD-L1 statusStrongly Positive vs. Not Strongly Positive

The registry-reported analysis text also identifies covariate adjustment and stratified analysis as concepts in the primary results. This is distinct from claiming that the model has removed every possible source of confounding. Randomization remains the fundamental basis for the treatment comparison.

16. Time-to-Event Analysis and Censoring

OS is a time-to-event endpoint because both the occurrence of death and the time from randomization to death contribute to the analysis. Participants who have not experienced the event by the relevant follow-up point can contribute partial information and are censored according to the registered definition.

Kaplan-Meier concept
S(t) = ∏ti ≤ t (1 − di/ni)

In a Kaplan-Meier estimator, di represents events at an event time and ni represents the number at risk immediately before that time. The estimator updates the estimated survival probability at each observed event time.

The ClinicalTrials.gov record does not report a Kaplan-Meier numerical reconstruction, median OS, or the number of OS events. Those quantities therefore are not added here. The registry does, however, provide the endpoint definition, censoring rule, and formal log-rank/Cox analyses.

17. Confidence Intervals and Precision

The primary OS estimates illustrate why a point estimate should never be read in isolation.

AnalysisHR95% CIWidth of CI
Initial OS0.820.67–1.010.34
Updated final OS0.800.65–0.980.33
PD-L1 ≥1% CPS OS0.740.58–0.930.35

The initial and updated OS estimates are numerically close, but the confidence intervals provide the important context for their precision. The PD-L1 ≥1% CPS analysis is a restricted population, so its estimate answers a different question from the all-participant primary analysis.

18. P-values and Superiority Testing

The registered hypothesis type is superiority. In this framework, the null hypothesis generally represents no treatment difference for the specified effect measure, while the alternative allows a treatment difference in the prespecified direction.

Initial OS

HR 0.82, 95% CI 0.67–1.01, P = 0.03160.

Updated final OS

HR 0.80, 95% CI 0.65–0.98, P = 0.01605.

The p-values should be interpreted within the registered analysis framework. They do not provide a measure of how large or clinically meaningful an effect is. A complete interpretation combines the p-value with the effect estimate, confidence interval, endpoint definition, analysis population, and study design.

19. Safety

The ClinicalTrials.gov record reports serious adverse events by arm. These counts are presented as affected participants over the corresponding at-risk populations.

GroupSerious adverse eventsAt risk
Pembrolizumab First Course110246
Standard Treatment First Course92234
Pembrolizumab Second Course02

The ClinicalTrials.gov record does not provide additional safety-effect estimates, confidence intervals, formal hypothesis tests, or definitions for the serious-adverse-event categories beyond the affected/at-risk counts shown above. Those quantities are therefore not inferred or calculated here.

Interpretation note: the second-course category contains 0 affected participants among 2 at risk. Because the denominator is very small, that count should not be compared with the larger first-course groups as if it represented a similarly informative randomized safety population.

20. What the Primary OS Results Do — and Do Not — Establish

Effect estimate

The initial OS HR of 0.82 and updated final OS HR of 0.80 both estimate lower instantaneous mortality in the pembrolizumab group relative to standard treatment under the reported Cox models.

Not an individual prediction

A hazard ratio is not a prediction of an individual participant's survival time or probability of death. It summarizes the relative event rate between groups within the statistical model.

Not an absolute effect

The HR does not directly communicate the absolute difference in survival probability at a particular time. Absolute time-specific survival measures, when available, would answer a different question. The ClinicalTrials.gov record does not provide those values.

Not a p-value substitute

The p-values of 0.03160 and 0.01605 quantify evidence against the relevant null hypothesis under the reported analyses. They do not rank the magnitude of treatment effects and should not be interpreted as probabilities that the null hypothesis is true.

21. Limitations

22. Why This Trial Matters Statistically

KEYNOTE-040 is a useful teaching example because the registry contains several layers of statistical analysis within a single randomized phase 3 trial: two primary overall-survival analyses, biomarker-restricted analyses, multiple progression endpoints, response analyses, stratified Cox regression, and risk-difference estimation.

ConceptHow it appears in KEYNOTE-040
RandomizationParticipants were randomized to pembrolizumab or standard treatment.
Time-to-event endpointsBoth registered primary endpoints were overall-survival analyses.
Log-rank testReported method for the primary and secondary time-to-event analyses.
Hazard ratioPrimary effect measure for OS and the posted progression endpoints.
Confidence interval95% two-sided intervals accompany the reported primary and secondary effect estimates.
Cox regressionUsed for treatment-effect estimation in the reported hazard-ratio analyses.
Stratified analysisPrimary OS models were stratified by ECOG PS, HPV status, and PD-L1 status.
Covariate adjustmentExplicitly identified in the primary analysis descriptions.
Risk differenceUsed for the two posted ORR analyses.
Biomarker-restricted analysisSeveral secondary analyses were restricted to PD-L1 ≥1% CPS.

23. Results in One Statistical View

EndpointPopulationEstimate95% CIP-value
Initial OSAll randomized participantsHR 0.820.67–1.010.03160
Updated final OSAll randomized participantsHR 0.800.65–0.980.01605
OSPD-L1 ≥1% CPSHR 0.740.58–0.930.00493
PFS per RECIST 1.1All randomized participantsHR 0.960.79–1.160.32504
PFS per RECIST 1.1PD-L1 ≥1% CPSHR 0.860.69–1.060.07736
ORR per RECIST 1.1All randomized participantsRisk difference 4.6-1.2 to 10.60.0610
ORR per RECIST 1.1PD-L1 ≥1% CPSRisk difference 7.50.6–14.60.0171
TTP per RECIST 1.1All randomized participantsHR 0.890.70–1.120.14545
TTP per RECIST 1.1PD-L1 ≥1% CPSHR 0.810.62–1.060.05851
PFS per modified RECISTAll randomized participantsHR 1.040.86–1.270.65759
PFS per modified RECIST 1.1PD-L1 ≥1% CPSHR 1.010.81–1.260.51982

The table illustrates an important statistical principle: the direction and magnitude of an estimate should be interpreted together with its confidence interval, endpoint definition, analysis population, and analysis method. The numbers are not interchangeable because the endpoints and populations differ.

24. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

25. Related Statistical Calculators

26. Sources

Continue with the statistical methods behind KEYNOTE-040

Explore the survival-analysis, inference, effect-measure, and clinical-trial concepts that appear in the registry's reported analyses.

27. Record Summary

KEYNOTE-040 provides a useful statistical case study in randomized time-to-event analysis. The registry reports two primary overall-survival analyses, both using log-rank testing and Cox regression, with the primary OS models stratified by ECOG performance status, HPV status, and PD-L1 status. The initial OS analysis reported an HR of 0.82 with a 95% CI of 0.67–1.01 and P = 0.03160. The updated final OS analysis reported an HR of 0.80 with a 95% CI of 0.65–0.98 and P = 0.01605.

The broader results illustrate why statistical interpretation should distinguish between endpoints, analysis populations, and effect measures. The registry reports hazard ratios for OS, PFS, and TTP, while ORR is reported using a difference in percentages. Confidence intervals describe precision, p-values address specified hypothesis tests, and stratification provides structure for the Cox analyses. None of these quantities should be interpreted in isolation.

Clinical Biostats methodology: A trial-results page should separate reported evidence from statistical interpretation. The purpose is to explain what the registered analyses estimate, how those estimates should be read, and what the ClinicalTrials.gov record does not establish.